Imejis.io
Give your agent a design studio to automate your marketing
86 followers
Give your agent a design studio to automate your marketing
86 followers
Your AI agent writes the copy. With Imejis, it ships the image too. Give Claude, ChatGPT, Cursor, and other AI agents a design studio via MCP. Create, edit, preview, and render production-ready graphics. Build OG images, social cards, marketing graphics, charts, certificates, QR codes and more. Design once. Generate at scale. Unlike generative image models, Imejis is deterministic: the same input produces the same pixels, keeping your brand, layout, typography, and data consistent. Free to start
Hi Product Hunt!
We're Dharmendra and Dharmik, twin brothers and the two people behind imejis.io.
Agents can write almost anything now. But ask one for a finished visual and you get a screenshot, a description, or an AI image with a garbled logo. We kept running into that, so we gave agents a real design tool.
imejis.io is a template-based image engine with an MCP server. Connect Claude, ChatGPT or Cursor to
https://api.imejis.io/api/mcp, sign in once, and your agent can create, edit, preview and render real designs:
OG images, social cards, certificates, QR codes and barcodes
22 chart types, including candlestick, treemap, sankey, funnel and KPI tiles
Permanent render URLs, so one template can make an image for every row of your data
It's a renderer, not a diffusion model. The same input gives the same pixels every time, so your fonts, logo and numbers come out right.
Not using agents? The same engine has a visual editor, a REST API, and Zapier, Make and n8n integrations. It already powers image generation for platforms like topmate.io.
Try this once you're connected: "Make a 1200×630 OG image for my post 'How MCP works' with our logo top-left, and export it as PNG."
100 free renders a month, no card needed.
For developers:
- A quick note on how two people ship all this: the backend (auth, workspaces, multi-tenant infra) runs on buildbase.app, our own SaaS backend platform, with Stripe for billing, Next.js on Vercel for the app, and Google Cloud for rendering at scale.
What would you have your agent render first? We're here all day to answer questions.
The thing that breaks when an agent makes images at volume is the font, not the layout. Where do the brand assets live?
@justin_rockmore Agreed - layout is deterministic, fonts are where it degrades quietly. A missing family doesn't throw; it falls back, and you find out when someone sees the post.
So fonts aren't something the agent passes per call. Each workspace has its own asset library, custom fonts uploaded once (one file per variant: regular, 700, italic), plus logos and images. sitting alongside the 500+ built-in families. Templates reference a family by name, and the agent only fills the dynamic fields. It can't choose a font, so it can't choose a wrong one.
The sharp edge we still have: reference a family that isn't in the workspace and it falls back silently instead of erroring.
@dharmendra_jagodana Thanks! Letting the agent fill dynamic fields is what makes this usable at volume, that's really cool.
Most of these tools hand the model the whole layout, and you find out when the client does (which usually look bad haha).
@justin_rockmore Thanks! Though I'll push back slightly on my own answer - the agent can design here, from a blank canvas, full layout control. The difference is when it happens.
Most tools regenerate the layout on every call, so every image is a fresh roll of the dice. Here the design gets authored once, a human opens it and checks it, and from that point it's an API - only the fields marked dynamic can move. Same layout, same type, call 1 or call 10,000.
So it's not that the model is kept away from design. It's that the design becomes a durable thing instead of a one-shot generation. That's what stops the "you find out when the client does" problem.